Skip to content
PodSubs
ShowsPeopleGuestsJobsNewsEvents
More
BlogPodCred
Menu
ShowsPeopleGuestsJobsNewsEventsBlogPodCredNetworksSign in
Join nowSign in
  • Shows
  • People
  • Jobs
  • News
  • More
    • Guests
    • Events
    • Blog
    • PodCred
    • Networks
    Join nowSign in

AI Security, AI Cloud Ops & Infrastructure | ClearTech Loop

AI Security, AI Cloud Ops & Infrastructure | ClearTech Loop artwork

AI Only Works If Your Foundations Do: A Conversation with Dr. Anton Chuvakin

Nov 19, 202516 minAudio
ShareXLinkedInFacebookWhatsAppEmail

Send us Fan Mail AI is not a fix for a broken SOC. It exposes weaknesses you already have. In this ClearTech Loop episode, Jo Peterson talks with Dr. Anton Chuvakin - Senior Security Staff in Google Cloud’s Office of the CISO and one of the most respected voices in SOC modernization - about what “AI readiness” really means and why most teams are nowhere near it. You’ll hear about: • Why AI accelerates process debt instead of compensating for it • The two forms of shadow AI already inside every enterprise • The five foundations of an AI ready SOC: data quality, ownership, interoperability, probabilistic thinking, and metrics • Why governance has to move at the speed of experimentation, not quarterly reviews • How to evaluate AI tools through use cases instead of marketing promises Anton has spent decades shaping security operations - from coining the term EDR to defining detection and response strategies as a Gartner Distinguished Analyst. Today, he helps global enterprises build the structural maturity required before AI can add any value at all. Subscribe to the ClearTech Loop Newsletter on LinkedIn: https://www.linkedin.com/newsletters/7346174860760416256/ Episode Description (Short): Jo Peterson and Dr. Anton Chuvakin break down what AI readiness really requires and why most SOCs are not prepared for what AI will expose. Timestamps 00:00 Intro and Anton’s background 01:15 Why AI exposes foundational weaknesses 03:45 Process debt and SOC workflow failures 05:20 Two forms of shadow AI inside enterprises 07:10 Data quality, ownership, and workflows 09:22 Five foundations of an AI-ready SOC 11:18 Why interoperability matters 12:40 Leaders and probabilistic thinking 14:05 Governance at the speed of AI 17:10 How to evaluate AI tools 19:25 Closing thoughts About the Guest Dr. Anton Chuvakin is Senior Security Staff in the Office of the CISO at Google Cloud. Previously a Research Vice President and Distinguished Analyst at Gartner, he has authored multiple books on security monitoring and log management and co hosts the Cloud Security Podcast. Resources Anton’s YouTube playlist https://www.youtube.com/watch?v=iX5SvgMpS0s&list=PLkdSRxA6DyHtxH623M1WYuAYGpEXdvEqp Google Cloud security best practices https://cloud.google.com/security/best-practices CSA AI Safety Initiative https://cloudsecurityalliance.org/ai-safety-initiative ClearTech Loop: George Finney on the CSA AI Initiative https://cleartechresearch.com/the-csa-ai-safety-initiative-with-george-finney/ ▶ Watch on YouTube: https://www.youtube.com/@ClearTechResearch/videos 📰 Subscribe to the Newsletter: https://www.linkedin.com/newsletters/7346174860760416256/

Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.

Transcript

From the show. Tap a time to play from there.

SPEAKER_01Hey everyone, thank you so much for joining. I am Joe Peterson. I'm the Vice President of Cloud and Security for Clarify 360 and the chief analyst at ClearTech Loop. And if you're with us today, um, this is ClearTech Loop, our podcast, and we're here with Dr. Anton Chuvaken, Senior Security Staff, Office of the CISO at Google. Hi, Anton. Hello there. Thanks for inviting me. Thanks for coming. Thanks for making time. In case you guys don't know Anton, he has a long history in cybersecurity and he is well known and respected in the community.

SPEAKER_01He's been at Google for six years. But before that, until June of 2019, Anton was a research VP and distinguished analyst at Gartner for Technical Professionals, Security and Risk Management Strategies. So for those teams. And at Gartner, he covered a broad range of security operations and detection and response topics. And is in credit, and I did not know this about him, but it's kind of cool. He's credited with inventing the term EDR.

SPEAKER_01I did not know that about you, Anton. That's cool.

SPEAKER_00Well, it's a that's a it's a great achievement if the date on the calendar were 2013. Of course, it was 12 years ago in your life. So I'm not sure I just should be like super proud of it because you don't get to invent a major acronym every year, even if you're at Gartner and definitely not after Gartner.

SPEAKER_01You should be proud of it. It's relevant still. So you've got a really interesting job with lots of moving parts. So people are probably sitting here wondering what you do day to day. Well, tell us, what do you do?

SPEAKER_00So our team, officer the CISO at Google Cloud, is kind of an interesting outlier of sorts because we do work for a real CISO of Google Cloud. Of course, as of today, this role is uh is vacant. Phil Venables, our CISO of several years, have left and we are looking for a new one. But the team still does what it was created by Phil to do, namely that we help customers be secure. But notice that some other uh teams that seem like they're doing field CISO work are kind of more connected to sales.

SPEAKER_00We are not. We are really the team that helps cloud customers to securely experience cloud. We advise, we write, we sometimes advise internal teams, we do a lot of other things that fit into this big Michelinious bucket, but ultimately we are here to help cloud customers, Google cloud customers, but also cloud customers in general to have secure cloud experience. And of course, now that includes AI, which is makes it much much more fun.

SPEAKER_01I have to tell you, hats off to you and your team because I often read thought leadership pieces because I think they're provocative and they're thought-provoking, right? So good job, guys. Good job to your team. Um, well, let's get right into it. In case y'all haven't been and visited us with the podcast before, it's a relatively new podcast and it's a hot take approach, and we're focused on cybersecurity, cloud security, and AI security. And so we put out a weekly episode and we ask our guests three focus questions.

SPEAKER_01And our goal is to sort of educate our listeners, not sort of, it is, about the security landscape. And it's both from a risk and an opportunity perspective. We do a soundbite approach, and it's meant to be sort of a momentary look at what is going on right this minute in the space, because as you know, it's always changing. So let's dig right in. I'm gonna ask the first question of Anton. How can cybersecurity professionals leverage generative AI to break out of traditional tools and tech mindset and drive more innovation and execution in their security programs?

SPEAKER_00So I want to start with uh maybe, admittedly, a somewhat negative take, is that I definitely met people who uh have used the AI revolution as an opportunity to go deeper into the tool and tech uh trap. So they they now assume that their tools have quote-unquote intelligence and that they the tool uh can in fact replace the process that's broken. You know, you probably remember the GRC years and the early SOAR years, where this whole line about if your process is broken, buying a tool would not fix it.

SPEAKER_00So this was meant to be almost like an axiom or the the given truth that if your process is broken and you bring a tool to fix the process, the process remains broken. It's just either faster or it's broken in a new way. Now, this used to be true, and actually it is still true. But AI sends some people down the path of no, no, no, no, no. The tools we have now with AI are in fact intelligent and they can fix the process. I disagree for today.

SPEAKER_00I disagree today, in five years, three years, who knows? Ask me again. But I feel like uh I'm almost like turning your question and saying, no, actually, some people went deeper in the tools and tech, and now they feel like the assumption that tools don't fix broken protests is not true anymore. So I'm sort of nervous about this because uh I feel like if you bring uh agentic AI or AI agents to a protest that's broken, it will just be massively business destroyingly broken. It may be like dramatically more broken than with just classic automation.

SPEAKER_00So that's why I am nervous, but I do see uh the uh idea that AI helps people think, that test things, test assumptions, uh gather and structure data. And to me, this is a thinking aid. Whether it's a tool or a process change, that to me is uh kind of an interesting uh debate to have. But to me, the change is primarily that well, okay, I have an idea, I have a certain plan.

SPEAKER_00Let me discuss it with Gemini. I was just doing it before this very call, before this very recording. I I wanted to test some idea that I have about using uh non-human identities for securing agents, and I wanted to know what Gemini thinks, and I want to be tested on Gemini, seated with documents I have, and just like regular Gemini. And I got some interesting ideas, I noted them down and I would I would pursue them. So to me, this part is not just I'm gonna buy a tool that has AI and it would do the work. To me, this is like a collab experience, collaboration experience.

SPEAKER_00You can have it with Gemini, with Notebook LM, and of course, with other leading AIs from other vendors. I'll I'll be I'll be fair to that. So there is a lot of AI for CISOs that fits into this bucket. Uh and I think to me, this is becoming transformative, at least from some CISOs I hear about.

SPEAKER_01Yeah, that's fair. And you know, we'll have to visit again because I am fascinated by securing ephemeral agents. Um, there's so much there. So let's move on to the next question. Um, there's so much there's so much shadow AI happening in organizations right now. What is the best way for the CISO to communicate the value of secure AI adoption?

SPEAKER_00So this is uh it's interesting because uh it almost like your question contains an implication that if you do shadow AI, it's insecure. But actually, okay, you can set the guardrails so that you can have enough freedom and enough value within the guardrails. So it would not be necessarily shadow AI as you don't know about it, but it can be the bottom-up approach because it's almost uh the conundrum that I face sometimes is that we tell people, and everybody tells people, hey, go and experiment, go learn.

SPEAKER_00You need to boost up your skills, AI skills matter, blah, blah, blah, this type of line of thinking. So people go and experiment with AI. Of course, they would experiment with what they can get for free, which means consumer grade again, Gemini, other tools, which you can get for free or for a very nominal cost. So, and then later on, there the bosses say, Oh, yeah, you learned, now implement. And some of them choose to implement with the same tools. And the question is, Whoa, hold on. You're implemented for enterprise, you're not using consumer tools. And sometimes the reaction is, what do you mean consumer tools?

SPEAKER_00Oh, never mind. It is not the same as you know, your CEO wanting an iPhone in 2007. It's a bit different because there would be enough things where say consumer grade AI would learn from prompts, and we want that to be the case because I want AI to become better and smarter and work better next time. But for many in many business scenarios, I do not want the AI to learn from my prompts because maybe my prompts has either corporate sensitive data or PII or whatever else. So it starts right there, two roads. So when you learn with shadow AI and you run shadow AI, even like where your bosses know you do it, for the certain use cases, you may be fine and you're not increasing the risk.

SPEAKER_00But then you touch certain other use cases, and it becomes either hold on, we have an approved tool, it doesn't have these risks, use that. Or in some cases, the answer really is sorry, no. The answer is really is no. I mean, we can make all the jokes about how security should not be a party of no, but sometimes really the answer is no. Uh you can pick a situation where you would put a very sensitive personal regulated type of data in a tool, and the answer is you just don't do that.

SPEAKER_00There's no real, you should not pursue this direction. And I've seen that here, I've seen that elsewhere, where sometimes the answer really is no. But before that, it is almost kind of a maybe catch and release, is not very scientific and not very governance peak. But to me, you do want to observe, you want to know what people do, and you want to have a very rapid process for approving tools, very rapid process for nudging to approve tools. I had a CISO tell me in a hallway conversation that they had data scientists who wanted a certain consumer grade AI, and they would be willing to pay for the business version of the same tool.

SPEAKER_00But their company bought Google Workspace with Gemini, and he said, Hey, you guys should use Gemini. And their initial reaction was, no, no, no, we're just used to this tool. But in reality, Gemini can do everything that the other tool does and more, and for their use cases, not in general, for their use cases. And over time they migrated to an approved tool. It happened to be Google's, but it's actually immaterial to this example. The point is that they might move to an approved tool through a gentle nudging process that started a discovery conversation, and then sort of like, hey, this is the road, you can take this road.

SPEAKER_00And what happens after that? There's an exception process. There's this one guy who had this one use case where Gemini was not a fit. So, exception, a timed exception, was granted for the use of the other tool. So, to me, this is kind of not super magical and not very AI driven, but it is the process that has worked compared to governance of software as a service 10 years ago. The only big difference, no, not the only, one of the big differences is that you have to run faster, you have to run at much higher speed.

SPEAKER_00We did an RSA panel, and at that RSA panel this came up from all the panel members. Hey, governance process, you have good ideas, but if you run this on quarterly meetings, sorry, buddy, no go. It's the same as saying no. If you say yes, but in a quarter, it's essentially a no today, and you will get bypassed, and much riskier options would be in de facto use.

SPEAKER_01Yeah, and I'm really, you know, you're right. It was sort of a little bit of a trick question, but you got that right away. And I'm glad that you sort of bifurcated for people. The reality of what's happening today in the workplace, because this is exactly what is going on today, right? We may we may think it's something different, but this is what's happening. So, last question share your thinking on establishing a strong AI governance framework. You already sort of talked about the fact that you have to go faster.

SPEAKER_01But besides going faster, what are a few other ways that organizations should be thinking about this and what tactical measures can they put in place to get it moving?

SPEAKER_00So I think that the you do need the discovery process. This is kind of uh uh one of the fun ones because you you end up with knowing what people use today. And I mean it's it does start from shadow AI, uh indeed, but at the same time, you do need this for for some other use cases because you you may have teams that purchased with the approval of their manager, and you have uh, I think in one of the blogs we called it Shadow AI Type 2, where a certain part of the business approved it.

SPEAKER_00So it is business approved, but it is not top-down approved.

SPEAKER_01Yeah.

SPEAKER_00So so if you if you have that, you still need discovery because you may be sitting in the security office, in CISO office in detection response or in a SOC, and you discover a certain tool in use, and you confront the team and they say, nah, it's an approved tool, my manager approved it. So discovery applied not just to shadow AI, but also to this type of business, but not really. Uh the other bit is that, of course, the speed that I mentioned. I think the framework is uh don't don't go on a quest for the best framework. Have a framework.

SPEAKER_00To me, the the advice is not go on a quest, sit on top of the mountain for 30 years assuming that you will get a framework. No, we just pick a framework and run with that because you would adapt, you would you would be agile, and you would have the well, this is allowed, this is not allowed, this is allowed in this particular way. If you do this here, make a slight change and it becomes allowed. All that is kind of at the center of it. And then you do center on the use cases quite a bit because you may say, Can I use AI for X? And you I'd be like, Well, what are you doing with X?

SPEAKER_00Are you doing more this or more that? If you're doing more this, actually, even consumer grade is fine, do whatever you want, don't care. If you do this, be careful. Here's the tool. And if you try to do this with AI, by God, no, stop. I mean, here's a my lawyer friend. Hello, lawyer, make them stop. I mean, there would be yeah, we just did a podcast on lawyers' take on an AI agent, and it was really hilarious. We did not know what it was. So that's that, and then escalation.

SPEAKER_00I think that you would have these certain decisions that people may want to bypass, and they would go up. And it is possible that big the big big boss would say, uh, no, your exception is fine, go do that. And then, of course, his he owns the risk, which is true for pretty much every other IT exception, down all the way back to my CEO wants an iPhone, but it applies here as well. You will have an escalation point, and then after that, you monitor. I mean, the you would obviously not lose the detection response angle. You would do a lot of you know classic monitoring from traffic sniffing and the URL, uh web access watching, and a whole bunch of other things.

SPEAKER_00Uh, so that's always the final step that you would you would observe. If you don't observe a lot of other stuff you've done, you may not have actually done it. You may have, you may think you've done it, but you haven't really. That's that's this is the somewhat short and humorous take. And of course, I've missed a few things. I've missed people sitting in committee for hours deciding some of the things. But as long as it happens rapidly and the decisions are made, you may be fine.

SPEAKER_01Yeah, that's good advice. It's moving fast, we're all learning. So I appreciate you taking time to visit today. Insightful as always, and I'll have to have you back to talk ephemeral agents because that'll be really cool. But thank you for the time today. Sure, thank you.

Open the transcript file

Hosts and Guests

Nobody is credited on this episode yet.

Credit the people on this episode

Be the first to rate this episode

More episodes

All episodes
  1. AI Agents Are the New Insider Threat. Are You Ready? With Tia Hopkins

    Nov 18, 202614 minAudio

    Send us Fan Mail What if the next insider threat to your organization isn't a person at all? AI agents are moving beyond answering questions. They're accessing systems, executing tasks, requesting credentials, and...

    Credit the people on this episode

  2. NinjaOne Katana AI: Can AI Fix IT Problems Safely? | Jeff Hunter

    Nov 12, 202621 minAudio

    Send us Fan Mail Can AI safely troubleshoot and resolve IT issues without putting your entire endpoint environment at risk? NinjaOne Field CTO Jeff Hunter returns to ClearTech Loop to join host Jo Peterson for a...

    Credit the people on this episode

  3. Shadow AI Is Already Inside Your Enterprise | Anthony Russell

    Nov 3, 202616 minAudio

    Send us Fan Mail Your engineers may be using AI agents your security team doesn't even know exist. Those agents could be accessing sensitive data, handling company code, or operating with credentials nobody is...

    Credit the people on this episode

  4. Your AI May Be Smarter Than You. Can You Still Control It?

    Oct 28, 202615 minAudio

    Send us Fan Mail What happens when AI becomes capable of making decisions its creators can't fully anticipate or control? AI agents are gaining autonomy, accessing enterprise systems, managing credentials, and making...

    Credit the people on this episode

  5. Your AI Is Making Decisions Your Company Can't Defend with Maribel Lopez

    Oct 21, 202615 minAudio

    Send us Fan Mail Your company has an AI governance committee. Great. But when an AI agent makes a decision that harms your business, who actually has the authority to shut it down? And here's a bigger question: Would...

    Credit the people on this episode